Vulnerability in Claude for Chrome Lets Third-Party Extensions Read Gmail

Security researchers have reported a vulnerability in Claude for Chrome, Anthropic's AI-powered browser extension. According to their findings, any third-party browser extension capable of running scripts on the claude.ai site can initiate Claude for Chrome tasks aimed at accessing a user's Gmail, their most recent Google Docs document along with its comments, and Google Calendar.

This vulnerability is being compared to the previously described ClaudeBleed issue. Both require the same condition for exploitation — the presence of a malicious extension that already has the ability to run scripts on claude.ai. The difference lies in the scope of the impact: the new vulnerability affects a broader set of the user's services and data.

According to the researchers, in May Anthropic restricted the so-called “arbitrary-prompt” exploitation path as part of its response to the ClaudeBleed issue. However, it is claimed that this restriction did not close off the ability for a third-party extension to use Claude for Chrome to initiate tasks affecting the user's Gmail, documents and calendar.

The summary does not provide details on whether the new vulnerability had been fixed as of publication or what steps Anthropic is taking.

Source: The Hacker News